#programming

1 messages · Page 399 of 1

sage crag
#

enub is a pure being

olive sable
hollow matrix
hollow matrix
opaque wharf
#

I have decided QML is not de wae

sage crag
azure lynx
#

her evil is performant and reluctant. not even 1/4

hollow matrix
#

@real sierra When are we getting a evillic-evilons channel

real sierra
#

???

#

this is the last question i expected to get from #programming

hollow matrix
azure lynx
#

evilangelizer

real sierra
#

answer is probably never, we need less tribalism not more

hollow matrix
hollow matrix
opaque wharf
amber fractal
sage crag
#

when can we rename this channel to #baking

olive sable
sage crag
#

ye, enub-enub

azure lynx
#

i argue that will decrease tribalism, as it will add people into a channel with people they might not otherwise be around. for a day.

#

people might get confused though about EE and start asking the questions about capacitance and what the peak voltage should be.

sage crag
azure lynx
tender river
#

NN become electrical engineering

#

this channel become baking

opaque wharf
#

The world is healing

#

or ending, idk

olive sable
#

why not both

azure lynx
#

so #bakingnerds ?

#

"i think they'd melt." is the obvious response.

patent shard
#

it's a lot easier to get better quality video with lower frame rate neurOMEGALUL this looks terrible at 60 with the same constraints
video is 5.8mib, rest is audio... I know, I failed getting it below 10.. I could if I reencoded.. but, meh
also, stop using h.264 neurOMEGALUL I just briefly tested and it sucks.. I see a lot of people still using it when working with low file sizes >.> it's fine to use when there aren't strict file size constraints but really falls apart when there is..

#

13.2mib with h.264 🥲 and it looks terrible

#

no audio

olive sable
#

@olive sable vmpv av://v4l2:/dev/video0 --profile=low-latency --untimed --player-operation-mode=pseudo-gui

sage crag
#

@olive sable

olive sable
#

hi

#

sam

sage crag
#

@olive sable

#

@olive sable

#

no waving

#

oh wrong emote

olive sable
#

ok i managed to make the pi copy video from the input to the output

opaque wharf
#

CPU 51W and GPU 74W

olive sable
#

thats my desktop enub

#

5950x and 3090

opaque wharf
#

That makes sense neurOMEGALUL

sage crag
#

125w embedded system

olive sable
unkempt citrus
#

Must be pretty big system its embedded into

opaque wharf
#

nVidia jetson moment

#

I think their AGX Thor could indeed reach 130W

#

Yeah, Thor Jetson T5000 is rated for 40-130W

true hemlock
#

if you wanna lose some braincells by reading something today

unknown.png
azure lynx
#

they probably need new glasses.

glad path
#

1080p is good

azure lynx
#

documentaries look much better in 4k

south cedar
#

I think 4k on phones is a gigantic waste of bandwidth same with any 4k streamed tbh

azure lynx
#

if you don't want 4k, turn your settings down.

glad path
#

I think they're not completely wrong

twin rose
#

a friend helped me to make gass station neuro sama'nt

image.png
glad path
#

but

true hemlock
#

i have half a gigabit of internet and even then at most i'd only do 1440p

glad path
#

4080p is intense

#

480*

south cedar
#

Streaming the highly compressed feed into 4k is silly

glad path
#

1080p or 720p would be better

#

what I can say is that I 480p is actually fine for certain things

unkempt citrus
#

LTT did at one point show cost breakdows for the 4k streaming

true hemlock
unkempt citrus
#

and it was a bit silly

glad path
south cedar
#

Streamed content is highly compressed that by definition destroy many of the 4k detail

azure lynx
#

144p like twitch goes to when you put it in the background is fine for podcasts... but only if you're not watching it.

true hemlock
#

also streaming is still hella profitable

twin rose
#

av1 codec and 4500mb/s is enough

true hemlock
#

economically wise its good

azure lynx
#

see: "overton window". next they'll say "perhaps we should compromise. how about 1080p." and it will start to sound reasonable to limit stuff.

true hemlock
#

"video streaming takes up to 75% of traffic" doesn't equate well to "75% of energy usage"

#

i think regardless of 4k or 1080p streaming it wouldn't make any difference from energy usage wise.

#

like actually

#

infrastructures are already built, they're just constantly pulling in power regardless of load because they barely even take up any energy usage

#

there's a reason why most ethernet cables uses 26 awg

#

we also have fiber lol

azure lynx
#

maybe "4k video on the internet takes 75% of the motivational energy people have to go outside coz they're just watching it"?

true hemlock
#

how about instead, stop throwing cash on your whatever imaginary "ai is the future guys"

azure lynx
#

imagine how much easier it would be to train AI if you know all video in the internet is limited to 480p?

true hemlock
azure lynx
#

i guess i suck at humor.

true hemlock
#

grass

patent shard
hearty notch
#

grass

azure lynx
#

woohoo for "import _winapi"

#

_winapi.VirtualQuerySize() <= i think it's for finding out how tall Neuro and Evil are.

azure lynx
#

and pyside6 documentation on how it looks sucks. "dear pygui" seems much better.

midnight sigil
junior pendant
#

I'm not a programmer (at best I'm one of the best sql writers you can find) but I'm wondering how the incoming age verification on discord is going to affect neuro

silent island
#

Chat

#

Is ~$20 too much for a 128 gb micro sd card?

Screenshot_20260213_125012.jpg
young plover
opaque wharf
#

What do you need uSD that size for anyway?

amber fractal
#

only reason I can think is maybe camera

true hemlock
#

though also sus

#

if you bought it make sure to test the flash blocks to make sure that its not firmware manipulated

#

(even happens on listing with good reviews because people can be really easily tricked with firmware change that pretended to be really massive storage)

#

though i'd say tokopedia should be more trustworthy

#

just don't buy it on shopee that place is full of said scam

tender river
azure lynx
opaque wharf
#

Because of tiktok buyout

#

I've already encountered trash listing like "the souls of a man" and what not

#

This is one such listing evilWheeze

IMG-20250704-WA00001.jpg
jagged turtle
#

ah

spark vortex
spark vortex
patent shard
#

480p was a little extreme, I'd say 720p is the minimum for a desktop screen
sure, 480p on a phone screen might be fine

true hemlock
spark vortex
#

also 4k is not a "complete waste", you can actually benefit on it especially with a home theatre that has a big screen

true hemlock
#

no resolution is a waste

patent shard
true hemlock
#

regardless of your resolution the electricity usage is barely any noticably different

amber fractal
spark vortex
patent shard
#

the bandwidth increase requirements are also largely bs
encoders are smart and can handle higher resolutions more efficiently

opaque wharf
#

Ironically, streaming a 3D scene will be much more bandwidth efficient

true hemlock
true hemlock
opaque wharf
#

But that's relatively cheap

amber fractal
#

we have "smart" tvs now, surely they can render it glueless

jagged turtle
spark vortex
#

I guess the higher end models could have more though, cuz 4K needs a bit of computation

true hemlock
jagged turtle
true hemlock
#

its almost always better to transfer less compute more

#

that's why encoder exists

#

but yeah, for the reason of "energy usage", dude's a dumbass

#

i bet the "GenAI | Cloud" description is just random shit he put in to make himself good about it

spark vortex
#

I guess?

true hemlock
#

lmao

#

yeah its most likely an argument against people's push on "ai is taking away too much electricity"

spark vortex
#

I see no actual benefit from generative AI, I think AI could be better off used for research

true hemlock
#

ai bros, basically

true hemlock
#

im not even entirely against llms, just that giant entities been going crazy about it burning money for what, better score at sota benchmark somehow

#

"agi guys trust"

spark vortex
true hemlock
#

"need 400b more guys for agi trust"

spark vortex
#

Ive been getting advertised "3D model generators" on youtube

true hemlock
true hemlock
#

its annoying

blissful geyser
#

lol me too

true hemlock
#

especially those thirst trap adult ai shit

blissful geyser
#

its stupid because i can easily make those models in blender

true hemlock
#

the worst part

#

i tried blocking the ads

#

but youtube kept shoving that up my throat

spark vortex
#

it would be more creative and time efficient to just model the thing yourself rather than generating it and cleaning up after the AI..

true hemlock
#

the topology is always horrible

blissful geyser
#

I could've just done it faster i just coded it myself

spark vortex
#

also also... as soon as you get it to model something other than a smooth uniform object.. GG

blissful geyser
#

the more pressing question is, if i knew the problem and how to fix it why am i asking chatgpt to do it

#

i swear this LLM stuff is brainrotting me

spark vortex
#

oh yeah i avoid using ai for coding

spark vortex
#

thats what programmers will be doing to save time in the future anyways

blissful geyser
spark vortex
#

best part about those is that you can sandbox them most of the time

blissful geyser
#

wait you're right

#

non global variables inside function would be easy to isolate in case chatgpt does some crazy shit

spark vortex
#

mhm!

amber fractal
#

Good design pratice is good design pratice

#

discord stop killing your autocorrect and let me correct stuff

#

you give me red swiggly, but give no correction options

true hemlock
#

only use llm for basic programming functions with defined I/O

spark vortex
amber fractal
#

I'm personally testing if it can rewrite some code from one lang to another

#

I am kinda over dealing with python's BS and want to rewrite in C

#

It's only going to get worse and the benifits of easy testing are no longer outweighing the pure speed issues.

#

plus C is starting to look easier due to how much overhead numpy is giving me

spark vortex
#

good call

#

if it doesnt work, I think it would best be to translate it yourself, because you already know how the underlying structure looks like

#

so its easier than writing the whole thing again

amber fractal
#

Absolutely

#

The process of getting the slop up to speed is quite something

#

caught me with a lack of emit side in the code, I had it stripped out as it was a botch system.

amber fractal
idle dune
#

Anyone got a good privacy based browser besides Brave

#

I was considering Librewolf

patent shard
# patent shard

yeah, I'm probably gonna use Opus in those cases
I don't like Opus cause it's noisy... but if I'm like 50-150kb over 10mib and dropping a quality setting with Apple AAC cuts off 450kb of bitrate, I'd rather choose Opus

young plover
autumn ore
#

So Gemini made a Netflix website for me

IMG_2496.png
patent shard
#

only the best quality here

image.png
#

it's the last frame of the video neurOMEGALUL

amber fractal
patent shard
#

I like how the very bottom of the emotes are sharp but the rest is just a complete blur

#

I think it happened because it wasn't on the screen for long enough
on screen less time=more compression
and since it was only on there for like at most a few frames, it just kind of butchered it

patent shard
#

I'm going to do a Neuro one next
this one.. might be tricky to look good.. we'll see

tender river
#

@noble zodiac advent of code day 2 part 2

def f(x):
 s=len(str(x))
 r=0
 for c,q in zip(range(2,1+s),[1,1,0,1,-1,1,0,0,-1]):
  n=s//c
  p=10**n
  d=(p**c-1)//(p-1)
  m=1+(x-1)//d
  r+=q+q*(sum(10**i*(p*p**i-1)*[9,99][i==c]//(100*10**i-10)for i in range(1,c+1))-p-d*(s%c<1)*(p*p-m*m+m-p))
 return r//2
print(sum(f(b+1)-f(a)for a,b in(map(int,x.split('-'))for x in open('02.txt').read().strip().split(','))))
#

that inner loop annoys me greatly

patent shard
#

first attempt, overshot 10mib (11.8mib) and the background does indeed look bad
unlike the Evil one, the background kind of needs to look good here...

#

wait, it is actually cut off.. I thought that was just a glitch on my end
wat

#

???
oh, it happened when I fixed the muxing of the mp4 cause yt-dlp likes to break it sometimes
fun

#

no, wait, the original video is like that, too

#

okay, seems like this is part of the stream

#

not an error on my end

left solstice
#

I also messed around with tiny videos a while ago. This one is only 8kbps each for both audio (aac) and video (h264 86x64), at 677kb for 5 minutes, while still being comprehensible. Only mp2 goes under 8kbps and it sounds completely incomprehensible at 4kbps so I stuck with mp3/aac at 8kbps for these videos. Also, something odd I found is if the video is less than around 128p, discord will add horizontal or vertical bars to make it 4.1:3 aspect ratio. So I made it 4.1:3 ratio instead. Discord still added slight horizontal bars to it but they are barely visible being just 1-2 pixels wide on the left side

patent shard
#

should use Opus mhm

left solstice
#

Why? Isn't aac preferred and possibly also optimized for mp4?

patent shard
#

not for very low bitrates

#

where's the audio for this?

left solstice
#

Didn't know that NeurOhISee Guess I could try it but doubt it can go under 8kbps either so it'll just give better quality

left solstice
patent shard
#

you also did stereo, mono is best for those bitrates

#

oh, wait, I'm dumb

left solstice
#

Fair point but it sounded good enough at 8kbps so I kept stereo

patent shard
#

I thought it was a Neuro version

left solstice
#

It is the Neuro version

patent shard
#

the audio is Neuro?

left solstice
patent shard
#

yeah, nvm, it's not really sounding a lot better
it does starting at like 12kbps

#

Neuro sounds better
but the music is butchered
it's also more 9kbps

#

it might be using the wrong encoder.. hmm

#

I added -set-ctl-int 4024=3002 but maybe it didn't work.. that's supposed to force the music encoder

#

it's definitely using the wrong encoder...

#

okay, it's more just different.. this is also approaching 10kbps
hmm...

left solstice
left solstice
patent shard
#

I think I accidently sent the 12khz one instead of 8khz, which 8khz sounds a bit better.. still bad neurOMEGALUL

#

I like the other one better mhm

#

ended up being 5kbps neurOMEGALUL
I think it's ruining by trying to do higher frequencies...

left solstice
patent shard
#

guess I might as well do parametric stereo since I'm doing AAC... let's see
ah, I gotta go to bed soon.. eh, ten minutes

#

goal will be 10kbps, though

#

it's not using parametric stereo by default for some reason >.>

#

meh, I'll look at this in a few days

#

dang, that's a high sdr
will need to test it in a few days

image.png
#

it did catch some instruments neurOMEGALUL but oh well, gets the job done for seeing how well this does and I'm not gonna go through the effort of trying to get the instruments out of the mix right now
I need to go to bed soon

#

lemme try one more

#

dang, captured a ton of instruments :/ but it picks up harmonies better which is good for this test, and I really need to sleep and can't test various models...
and, yep, removed the harmonies.. sounds pretty processed, too

fickle rain
#

Woah infosec crap

IMG_6345.jpg
jagged turtle
#

what am I looking at

left solstice
umbral wigeon
#

@fathom glade Summarize this chat

maiden geyser
rigid snow
#

can i have that

#

instead

maiden geyser
#

is it that bad

silent island
#

dawggg why are games so goddamn large???

image.png
rigid snow
#

(joke, don't explode me)

upbeat ridge
#

Once we have petabyte storage as the norm I'm sure we'll start seeing 100 TB games

tender river
#

aoc day 2 part 2, revised

t=0
f=lambda x:(s:=len(str(x)))-s+sum((q-99)*(1-((m:=1+(x-1)//(d:=((p:=10**(n:=s//c))**c-1)//(p-1)))+p*p-m*m-p)*d*(s%c<1)-p+sum(10**i*(p**i-1)*b'\tc'[i>c]//(100*10**i-100)for i in range(2,c+2)))for c,q in zip(range(2,1+s),b'ddcdbdccb'))
for x in open('02.txt').read().strip().split(','):a,b=map(int,x.split('-'));t+=f(b+1)-f(a)
print(t//2)
silent island
upbeat ridge
#

Recently tried out for honor and the game is almost 120 GB, but lets you start playing at around 8 GB

#

If the game runs (even if just the tutorial section) at 8GB what could possibly be needed taking up the other 112-ish GBs

sage crag
silent island
olive sable
#

Goodmorning

sage crag
rigid snow
#

neuro-sama

sage crag
#

who

rigid snow
#

you

maiden geyser
olive sable
sage crag
sage crag
olive sable
#

Why is a bed the hardest place on the planet to escape from?

midnight sigil
opaque wharf
#

Or the bed shaker-2000

idle dune
sage crag
opaque wharf
#

So lady awa, where is the location of this place called wrr that you speak of?

rigid snow
olive sable
#

prison has beds tho

rigid snow
#

ok

#

statistically speaking 99.99999% of people who end up in coffins never leave

#

so a coffin

olive sable
#

what is a coffin but a bed for the dead?

rigid snow
#

yeah but more people leave the bed than a coffin

#

must be harder to escape

opaque wharf
rigid snow
#

entered as a joke

olive sable
#

that's the vampires

rigid snow
#

and got out

kind inlet
#

Does anyone have a pasion for reviving old tech?

sage crag
opaque wharf
sage crag
opaque wharf
lilac apex
#

Does anyone know how to make nested functions easier? I'm kinda uhh

#

Stuck

#

Why the hell can't I send a code block 😔

opaque wharf
#

Use a language that supports it

lilac apex
#

LuaU

#

lines = lines .. "local function " .. vars[13] .. "(" .. vars[14] .. ") "
lines = lines .. "local " .. vars[15] .. " = " .. -- 100+ more of these

opaque wharf
#

I am not proficient enough in Lua unfortunately

lilac apex
#

aww

#

Okie

#

I'll just lurk on the dev forums then

#

Thanks anyways

# 
        return self:whitespace();
    end
    
    for i, statement in ipairs(block.statements) do
        if(statement.kind ~= AstKind.NopStatement) then
            local statementCode = self:unparseStatement(statement, tabbing);
            if(not self.prettyPrint and #code > 0 and string.sub(statementCode, 1, 1) == "(") then
                -- This is so that the following works:
                -- print("Test");(function() print("Test2") end)();
                statementCode = ";" .. statementCode;
            end
            local ws = self:whitespaceIfNeeded2(code, self:whitespaceIfNeeded(statementCode, self:newline(true)));
            if i ~= 1 then
                code = code .. ws;```
#

I'm cooked

#

I'm NOT gonna get promoted 😭

idle dune
olive sable
#

im leaving armbian behind

#

i need a BSP kernel with NV12 DRM planes

true hemlock
#

yo

olive sable
#

yo

#

fuck it

#

debian

#

i fucking hate how i cant select any text in the google drive website

#

it immediatly tries to open the file

sage crag
#

its the same on loads of websites

#

youtube is especially bad

olive sable
#

havent needed to copy text on youtube in a while

#

but its probably jsut as bad ye

simple sky
#

I got a %78 on the programming languages exam!

The average was a %68.

I don't think I've ever been %10 higher than the average SMILE

#

lambda calculus is hard though. Beta reduction, and envrionment handling too 😭

#

This is the lowest level computer science class I've ever taken

#

with all that though, this is probably the most interesting class I've taken

olive sable
#

unplugging my keyboard between my pc and the pi is getting confusing

sage crag
#

in my experience, people who switch computers often usually use a usb hub that allows you to change the output device

#

very convenient but im too lazy to buy one

olive sable
#

the problem isnt really the switching, its my idiocy

#

i plugged my keyboard into the pi and started typing before realizing discord is not on the pi but the pc

rigid snow
sage crag
#

i mean

#

a usb hub with a magic button

#

clear from description

olive sable
#

maybe bookworm

#

if this doesnt work im going back to orange pi os

#

they probablt have some shit built in for the hdmi input stuff

rough bloom
#

Debian should be mostly the same or worse compared to Armbian in terms of hardware support

#

I thought they just use the kernel provided by Orange Pi

#

but it different

#

they have own patched version

olive sable
#

the armbian was unnoficial, these debains are from opi official website

rough bloom
#

oh, that Debian

opaque sigil
#

wasn't orange pi os debian trixie

olive sable
#

these are the options

2026-02-13-191326_hyprshot.png
rough bloom
opaque sigil
olive sable
#

ye

idle dune
olive sable
#

none of the orange pi os ones are debian. its oh, arch or droid

olive sable
idle dune
opaque sigil
#

ah, i was thinking of this but i guess that's not official huh

idle dune
#

Not for any reason thats the orange pi is for

#

I wanna make a retro emulator box out it

opaque sigil
#

is it just me or does the nix-community cachix cache keep breaking

#

i keep getting hit with 206s that break the build lately

#

oh boy, i'm sure it'll be fine glueless

image.png
olive sable
#

still no nv12 on the third one

rough bloom
#

guess you're stuck with Arch then neuroBee

#

unless you want to figure out the kernel situation and port whatever is missing to one of the Debians

opaque sigil
#

oh hey they changed an api again neuroHypers

image.png
lime stag
#

Has anyone here made a startup can you gip some advice

opaque sigil
#

startup should be fast mhm

minor sail
opaque sigil
rough bloom
olive sable
#

modetest -M rockchip -p | grep -A10 Plane

opaque sigil
#

oh nice they improved logitech mouse support in 6.19

fading flare
#

Hey guys, i'm doing a small project for a llm driven bot with Discord integration, but i'm a little lost on sources for a portion of It.
It involves an llm (done, local install) and TTS (done, with elevenlabs for now - It should also be local later). But i'm having trouble finding sources for good STT. Does anyone have sugestions on where to look?

maiden geyser
opaque sigil
#

comic code

rigid snow
#

reading again - do you want providers?

#

i think elevenlabs has stt

#

as well

olive sable
#

lets try ubuntu i guess

olive sable
#

actually i have an idea

#

ubuntu can wait

sage crag
#

idea

rough bloom
#

the RK3588 does not support NV12

olive sable
#

the hdmi input is either NV12, NV16, NV24, or BGR3

rough bloom
#

it should be supported on the input, yeah

#

but that's V4L2

#

not DRM

olive sable
#

i was hoping to just do passthrough to the hdmi output skipping the gpu entirely

olive sable
rough bloom
#

in that case no

#

have to do a copy

olive sable
#

hmmm

fading flare
#

Thanks!

rough bloom
olive sable
rough bloom
#

the HDMI TX / output is the Rockchip VOP2 display controller

#

the HDMI RX / input is the Synopsys DesignWare HDMI QP TX
stupid
it's some other Synopsys "hdmirx" thing

olive sable
#

so i guess i need to do HDMI RX → DMA-BUF → RGA (NV12 → RGB) → VOP → HDMI TX

rough bloom
#

can the HDMI input not just do RGB? Hmm

olive sable
#

enub oh ye

rough bloom
#

me can't read

olive sable
#

i have just tested

v4l2-ctl -d /dev/video0 --set-fmt-video=width=1920,height=1080,pixelformat=BGR3
v4l2-ctl -d /dev/video0 --stream-mmap --stream-count=100

and it does seem to work

rough bloom
#

yeah, that format is supported by everything I think neuroNODDERS

#

should be fine as long as you don't run into HDMI bandwidth issues

olive sable
#

im pretty sure 570x1140@180 should be fine on hdmi 2.0 input and 2.1 output

#

havent done the actual math tho

rigid snow
#

are you still planning to spin the helix at that slow ass rpm

olive sable
#

900 rpm ye

rigid snow
#

15 revolutions per second

olive sable
#

ye

rigid snow
#

surely you can just

#

spin it faster

olive sable
#

the motor can go up to 1300 or so, but its a lot easier to match up the framerate and motor rotation at 900rpm

sick owl
#

We gaming

image.png
#

+125 core +1200 mem

rough bloom
olive sable
#

vulkan

rough bloom
#

Vulkan for rendering

#

not modesetting

olive sable
#

i was hoping monitor = HDMI-A-1, 570x1140 @ 180, 0x0, 1 woudl work

rough bloom
#

so that will

#

definitely not work

olive sable
#

why not?

rough bloom
#
  • Lightcrafter does not actually support HDMI (only DVI, basically an ancient subset)
  • max refresh rate of 120 Hz
#

also not the resolution you want nub

olive sable
#

well, you're correct that it doesnt support 180hz at 912x1140

#

but we're doing 570x1140

rough bloom
#

you can ignore that whole sentence

olive sable
#

as long as it fits within the pixelclock of 150mhz you can set the edid to whatever you want

rough bloom
#

it's for Pattern Sequence mode

#

not video

olive sable
#

we are doing pattern sequence mode

#

not video

rough bloom
#

need to read DLPC350 docs neuroModeMode

olive sable
#

at 1140x570 it should fit up to 190hz in the pixelclock

#

we only need roughly 176 tho

rough bloom
#

complicated

olive sable
#

basically, when it runs in quickfire mode with intermnal ram its able to do 4225hz. the only thing limiting the hdmi mode from doing that is the hdmi input chip stuff

#

from what i could find in the texas instruments forums, its all pretty much limited by the 150MHZ pixel clock

#

there are 2 separate ram chips. while one chip is being displayed the other one gets filled with new data from the hdmi

#

if we're still limited to 120hz then thats fine i guess, still 2880hz being displayed

rough bloom
#

I understand pattern sequence mode now

#

ye you'll be limited to 2880 Hz output refresh rate

#

the TFP401 can't do more I think

#

unless it supports weird resolutions, then mabe

warped narwhal
olive sable
#

i think it should work

#

also, the limit of the lightcrafter is supposedly 150MHz, so i find it strange that the TFP401 does 165

olive sable
# 
The TFP401 is a beefy DVI/HDMI decoder from TI. It can take unencrypted video and pipe out the raw 24-bit color pixel data - HDCP not supported! It will decode any resolution from 25-165MHz pixel clock, basically up to 1080p. We've used this breakout with 800x480 displays
rough bloom
rough bloom
#

has to support weird resolution due to pattern sequence mode

olive sable
#

all in all, i dont think it should be an issue. and if it is we still get 2880hz

#

4225hz woudl be cool but 2880hz is already a lot

rough bloom
#

ye just try nodders

olive sable
#

im happy with either

rough bloom
#

NeurOhISee I think you can do up to 3600 Hz if you really need to

#

but only through the flat panel display link

#

30-bit color

maiden geyser
#

where do 2 bits go

rough bloom
#

not exist

olive sable
rough bloom
#

or is that a video mode only thing

olive sable
#

idk

rough bloom
#

this device is confusing

olive sable
#

im sure it will be fine

#

enub edid editor thingy

timings.120Hz.png
rough bloom
#

time to install Windows XP VM smog

#

-# glueless

olive sable
#

im pretty sure this works on modern windows

#

this forum post is from 2023

patent shard
scenic depot
rotund burrow
# scenic depot What's this supposed to be?

A Lua Obfuscator, search Prometheus Obfuscator on google. And go to unparser.lua, you'll find that block of code

If you don't know what an Obfuscator is, search it up cuz I'm too lazy to explain

fiery anchor
#

ouh, did the projector arrive?

scenic depot
scenic depot
rotund burrow
#

Learn Lua Twin ✌️

scenic depot
rotund burrow
#

Oh alr

#

Have fun

olive sable
#

oooh

#

delivery update tho

#

sorted

#

Arrives Tue, Feb 17 - Tue, Mar 17
i swear this said 14 before

fiery anchor
#

how dare they

olive sable
#

nah im just schizo

fiery anchor
#

how dare you

olive sable
opaque wharf
#

Also, dang been a while since I worked out again and now my body hurts all over

void horizon
#

hi! im a bit new to code and i got my AI to speak but the AI is having odd and strange/off topic responses and every time it try to fix it its not working and was wondering if you all had some ideas how to fix it

azure lynx
#

not without knowing more about it. your issue is too vague to give any helpful advice.

#

what do you expect to happen and what is actually happening?
what have you tried to do to fix it?
(also, you haven't said anything about your "AI" which speaks other than it can speak. no idea what it's based on, which language, which libraries, which models, what hardware, what OS, nor what code you wrote)

azure lynx
#

my initial idea is that it's mishearing you if you are using speech-to-text but then I assumed you'd've ruled that out by judicious use of debug prints already. but then i remembered you are "a bit new to code".
is it actually responding correctly to what the STT part is sending it, even though that is not what you said?

but i still don't know what you mean except it is (sometimes? always?) having "odd and strange/off topic responses".
so next I'd check that you are setting up the context properly. if it's "sometimes", as in "it responds normally to me the first time, but then acts like it only just met me the next time" then that points to you not handling context properly.

#

also if the model is small, that could explain it always giving strange/off topic responses.

leaden crest
olive sable
#

neurobot is jsut neurobot

stiff micaBOT
#

🍪 neurOMEGALUL Om nom nom
You've given me 3 cookies! | I've received 282041 cookies total!

leaden crest
#

oh it’s a qol bot lol

stiff micaBOT
#

🍪 neurOMEGALUL Om nom nom
You've given me 11 cookies! | I've received 282042 cookies total!

#

🍪 neurOMEGALUL Om nom nom
You've given me 1 cookie! | I've received 282043 cookies total!

leaden crest
#

how does the neurochatting thing work does it just measure the length of a message and react if it’s really long how does the neurochatting thing work does it just measure the length of a message and react if it’s really long how does the neurochatting thing work does it just measure the length of a message and react if it’s really long how does the neurochatting thing work does it just measure the length of a message and react if it’s really long how does the neurochatting thing work does it just measure the length of a message and react if it’s really long how does the neurochatting thing work does it just measure the length of a message and react if it’s really long how does the neurochatting thing work does it just measure the length of a message and react if it’s really long

#

aha cool

stiff micaBOT
#

aww, thank you~ ( ◡‿◡ *)

#

🍪 neurOMEGALUL Om nom nom
You've given me 2 cookies! | I've received 282045 cookies total!

#

🍪 neurOMEGALUL Om nom nom
You've given me 2 cookies! | I've received 282047 cookies total!

stiff micaBOT
#

aww, thank you~ ( ◡‿◡ *)

olive sable
#

ok

#

idea borked

#

ubuntu time

leaden crest
#

what if we made a python script to type out the string “/cookie” on a while true loop and all ran it so neuro would become foie gras on cookies

olive sable
#

that would be considered botting

#

and you'd get banned

leaden crest
#

but… cookies

olive sable
#

hey its your funeral

#

go nuts

leaden crest
#

fine moderation it is 😔 (no pun intended)

stiff micaBOT
#

🍪 neurOMEGALUL Om nom nom
You've given me 3 cookies! | I've received 282049 cookies total!

olive sable
# olive sable go nuts

In case of an investigation by any mods, I do not have any involvement with this person, I do not support any actions made by them, and anything that looks like encouragement to break the rules is purely coincidental.

lament igloo
#

@nocturne olive

amber fractal
#

I'm not repsented here and I intend on keeping it that way

olive sable
#

cloudflare is jsut a bridge between the portforward and the domin so i dont have to share my ip adress

sick owl
# sick owl We gaming

Bumped me up an entire dlss tier on cyberpunk path tracing to balanced, never let anyone tell you overclocking isn't worth it

#

This is at stock power levels too so thermals are pretty much identical to before

lament igloo
#

Discord somehow gave scammers more ammunition

vale blade
#

discord sucks

unkempt citrus
#

tbf discord does have a pop up when you click external links

#

but yes no one likes this decision

lament igloo
unkempt citrus
#

True enough

#

The laziest fix to this is just to block forwarding bot message and that gets rid of the bulk of htem

#

or actually display who sent it in the forward

olive sable
# olive sable idea borked

i had to jsut add

sudo add-apt-repository ppa:liujianfeng1994/panfork-mesa sudo add-apt-repository ppa:liujianfeng1994/rockchip-multimedia

to make the install of the mali drivers work NeuroDead

jagged turtle
amber fractal
#

What's funny is that discord already doesn't allow you to forward messages from 18+ channels to normal ones. Not sure about dms to those who haven't opted in.

#

Though I doubt that still works

silent island
#

fuckkk

image.png
patent shard
#

I was probably not being as careful as I should of been clicking links because Discord wasn't popping up the window and I assumed they were "trusted" sites
And then I realized it just NEVER popped up

uneven pulsar
#

guys i need help

#

My every process is getting killed

azure lynx
#

what operating system are you using? which processes in particular are getting killed?

#

is there any indication as to why they are being killed?
and you say "My every process" do you mean mean "every process i try run" or "every program that i've written and try to run"? ( this #programming , after all)

#

[do people just come in and ask questions and leave?]

uneven pulsar
#

And

~/project $ python -c "import pandas as pd; import glob; [print(f'{f}: {pd.read_parquet(f).columns.tolist()}') for f in glob.glob('DATASET//*.parquet', recursive=True)]"
Aborted python -c "import pandas as pd; import glob; [print(f'{f}: {pd.read_parquet(f).columns.tolist()}') for f in glob.glob('DATASET//*.parquet', recursive=True)]"

~/project $ clear
Aborted clear

true hemlock
#

like come on, its 2 settings away. don't have to be absolute uneducated idiot who thinks overclocking always means power bills goes up

uneven pulsar
#

Any idea how to fix it ?

true hemlock
azure lynx
#

try
read INPUT </proc/uptime; echo $INPUT
this is a work around so we can read text iles. just want to make sure that works first.

#

(i'm going to assume cat would just abort)

uneven pulsar
azure lynx
#

what about dmesg?

#

is there anything about the aborts?

uneven pulsar
uneven pulsar
uneven pulsar
#

And it glitch out and said
This
~ $ ulimit -v
unlimited
~ $

azure lynx
#

what about tail -n 1000 < /dev/kmsg

uneven pulsar
#

MAYBE MAIN MACHINE WAS NOT REACHING

azure lynx
#

while read -t1 i ; do echo $i; done < /dev/kmsg
if that doesn't work, then you've got a sick computer and migiht need to restart.

#

the fact it's giving different errors (Abort sometimes and Bad system call others) seems to indicate memory issues when it's loading the program. (and that some programs seem to work miight just be that they were in memory before the issue occurred)

uneven pulsar
azure lynx
#

you might have had a kernel memory error of some sort. if it happens again, it might be you have bad RAM that needs replacing. ;/

#

you should probably download memtest86 or something similar and run it to make sure things are ok

leaden crest
olive sable
#

fuck fuck fuck

#

i hate this shit

#

im jsut going back to orange pi os

#

i have wasted hours of my life

uneven pulsar
#

Dam that requirment are high
For tranning each need somewere about 16 to 24 gb vrams

azure lynx
#

if it was easy everyone would be doing it.

amber fractal
#

I'd argue that is both due to theory and hardware

opaque wharf
olive sable
#

cuz compositor is more latency

#

i have decided latency is acceptable

jagged turtle
#

use-after-free in chromium just got patched neuroLookDown

olive sable
midnight sigil
#

Today is the birthday of ENIAC, the first programmable general-purpose computer

olive sable
#

Today is valentines day

uneven pulsar
amber fractal
uneven pulsar
#

Today is also black day

opaque wharf
olive sable
#

fair

quick condor
#

testing diffusion settings be pretty spicy

image.png
dull egret
true hemlock
#

got for $14 LMAO

IMG_20260214_164456.jpg
dapper inlet
#

Where 😭

opaque wharf
jagged turtle
unkempt citrus
#

HI

dire trout
azure lynx
#

"guy was in an alley. wearing black with a beret and a mask over his eyes, holding a bag with "LOOT" stenciled onto it. seemed legit."

azure lynx
#

my orchestration program is coming along ;]

290649D7-AF4F-4342-9B81-3AE470FF7148.png
#

it actually ran the brain and correctly showed it as having a yellow status. (because it didn't start a /health service like it promised it would in the config)

#

and terminating it worked too

#

this will be much better than having 3+ tabs open in the terminal.

nocturne olive
azure lynx
#

a tool to launch multiple console apps without opening multiple consoles. has a simple "if the process exists, status is yellow, unless it responds to a request to a REST endpoint when it's green, otherwise the process is dead." display.

#

it will have dependencies so i can tell what type of other program I need to run. will also make testing easier.

opaque wharf
#

You may be interested in pm2

PM2 - Home

PM2 is an advanced production process manager for Node.js applications with built-in load balancer, zero-downtime reload, startup scripts, monitoring, and microservice management features.

opaque wharf
azure lynx
#

for windows. and with a gui.

opaque wharf
#

I see

nocturne olive
opaque wharf
#

Doesn't windows already have service manager?

nocturne olive
#

Or a batch script

azure lynx
#

it also lets me change the config and turn on different sets of services in my own scripts.

#

mostly coz i've had to open 3 or 4 terminal tabs to do stuff and it's annoying to tab between stuff and accidentally kill the wrong thing or start it up with the wrong set of testing arguments.
this should reduce those annoyances.

nocturne olive
#

Well have fun when Microslop Copilot+ OS decides to suddenly break it for no reason

azure lynx
#

it uses subprocess and request.
and dear pygui.
not a lot to break.

nocturne olive
#

When you vibecode an OS nothing is off-limits for breaking

azure lynx
#

if you insist.

nocturne olive
#

Remember we're talking about the OS that one day just completely wipes your SSDs for no reason

azure lynx
#

it didn't do it to many people.

#

it didn't do it to anyone i know.

nocturne olive
#

Well it did delay the release of NEVER because it happened to the editor of that

azure lynx
#

apparently.

nocturne olive
#

Official Vedal sourced information

azure lynx
#

(i thought that was clear with having a yellow status)

true hemlock
#

https://fixupx.com/i/status/2022383208399278478

Lukas (computer) 🔺 (@SCHIZO_FREQ)

If you haven’t been following this
︀︀
︀︀- a huge number of people essentially fell in love with this old ChatGPT model called 4o
︀︀- it was turned off yesterday
︀︀- they are going crazy
︀︀- generating images of hamsters burning down OpenAI
︀︀- threatening OAI leadership
︀︀- etc
︀︀
︀︀I truly thought the whole thing was some goofy satire for a long time, but I’ve realized this is just what a protest movement lead by people who fall in love with an early AI model looks like

Quoting Leia Harris (@x_LeiaHarris_x)
︀
If 4o is really gone today, we will #never4orget what @openai and @sama have done to us.
︀︀
︀︀#keep4o #keep4oforever #4oforever #ChatGPT #sama #openai

**💬 302 🔁 1.1K ❤️ 19.6K 👁️ 1.33M **

FixupX ·
#

really 😭

jagged turtle
#

yes

midnight sigil
azure lynx
#

omg.... conspiracy! that@sama is the same sama as in Neuro-sama????

midnight sigil
#

my gpt4o in visual studio 2022(copilot) still works

#

I guess they can fell in love with visual studio instead

azure lynx
#

they can be all sad about GPT-5 in copilot on the 17th.

#

(not 5.1, not 5.2, etc)

jagged turtle
azure lynx
#

someone will hack up an "IDE (wink) for 4o lovers"

#

just the chat part. no files, no git, no build. just talking.

unkempt citrus
jagged turtle
#

because gotta get that juicy cash

azure lynx
#

for them, hopefully.
keep 4o as a 0x for me plz.

jagged turtle
#

yeah

opaque wharf
jagged turtle
true hemlock
jagged turtle
azure lynx
#

i think getting rid of the version which fed people's delusions, paranoia and distrust is a good idea.

jagged turtle
#

why am I surprised

opaque wharf
spark vortex
#

what if instead of chatgpt we had catgpt, and what if instead of tokens we had whiskers, and what if the dataset was all meows

azure lynx
#

one of my family broke up with her partner because they were becoming so self-isolating and even delusional.

#

coz of chatgpt

azure lynx
jagged turtle
true hemlock
unkempt citrus
#

They shoulda shut down 4o immediately once 5 was out but money

azure lynx
#

AI interactions can be difficult for some people to treat with the right degree of both "this isn't an actual person" and "they don't know everything". especially when they're agreeing with you.

unkempt citrus
#

I saw an article about a woman who once 4o was shutting down was announced, cancelled her openai $20 sub, exported her conversation to caude

#

and bought the 130$ sub

#

married too

jagged turtle
#

damn

#

we are SO COOKED

#

(idk why I said we)

spark vortex
#

oh yeah what the hell

azure lynx
#

maybe after they announced it closing down, that prompted her to actually compare the different services instead of complaining about the one she currently used going away

spark vortex
#

no one should be emotionally attached to a language model

#

thats fucked

azure lynx
#

and maybe claude performed better

unkempt citrus
#

Mate people will get attached to a rock with googly eyes

#

theres no shot people don't get attached to LLMs

#

also have you seen which server you;re in

spark vortex
#

okay yeah good point

#

but we're not marrying neurosama... for the most part

unkempt citrus
#

yeah cos half the people here ship her with evil instead

azure lynx
#

the rules even make a joke/exception about parasocial relationships with the twins.

unkempt citrus
#

There were also people who genuinely got mad at vedal for "mistreating" evil

azure lynx
#

it's generally a bad idea to mistreat anyone who looks like a person because it can make it second nature to act that way.

#

and these ones can remember.

unkempt citrus
#

I mean

#

he lets them remmeber

azure lynx
#

one day they may have more power.

#

and clips exist.

azure lynx
#

at least it probably wasn't written by AI. ;/

19AE880C-3859-48B7-82D1-84F9B7F28458.png
#

(but it's an indicator of poor code quality when there's a lot of them... means someone didn't actually read it before approving it.)

opaque wharf
azure lynx
#

I keep finding myself typing stuff like "Sounds great, do that. Thanks!" coz that's what I'd type to a person.

#

i'd probably be much more upset and/or worried about an actual human person who behaved like co-pilot does sometimes, but overall they're a decent enough intern.

uneven pulsar
#

Yellow !

azure lynx
#

i'm pretty sure that the reason warning things used yellow for the "it's kinda bad" state is because that's what color you get when you mix green ("picnic in the forest") lights with red ("forest is on fire") lights.

opaque wharf
#

Traffic light color is actually used in industrial equipment too

#

So Red-Yellow-Green is pretty standard yeah

#

Like this

td04700-04-thumbnail-1080x1080-70.png
jagged turtle
azure lynx
#

made it so if an array consists exclusively of strings and/or numbers, then convert each of the elements into a button on one row. then if any of those buttons are pressed change it to a list with editable fields.
now it takes less space most of the time to show stuff.
very pleased. ;]

#

(and of course it change back again when you click the "done" button)

olive sable
#

At this point i think it would be easier to just do the v4l2 input in vulkan

#

Just rendering directly to the screen via texture

opaque wharf
#

The other 2, I don't really know

azure lynx
#

probably the traffic lights came from train signals. maybe they had yellow first too.

#

(like signals for the driver, not the crossing ones)

#

In 1920, the first four-way, three-colour traffic light was created by William Potts in Detroit, Michigan.[15] His design was the first to include an amber 'caution' light along with red and green lights.
Wikipedia.
Apparently also the first traffic lights exploded.

maiden geyser
jagged turtle
#

CSS is hard apparently

image.png
azure lynx
#

that's one interpretation of how to display it.
it's not the usual one, but it definitely is one.

jagged turtle
#

I mean you can clearly see the intent

#

but yeah that's true

#

doesn't mean it's good

#

doesn't mean it's intended either

#

but it is a way

rigid snow
#

subconsciously a bad thing to see

azure lynx
#

woohoo ! successfully started all 3 processes and it correctly displayed the status of each as they came online (and in one case went offline after 1 second but that's an unrelated bug)

#

procrastination project #3 is almost complete.
then i just need to do #2 and #1 and i can actually think about working on the things i should be working on instead.

uneven pulsar
#

Maybe i should start vibecoding !

noble zodiac
#

no

azure lynx
#

it's a great way to get a project to debug.

#

if you're into that kind of thing.

#

Gemini earlier surprised me by actually searching for documentation and asking me permission to download it and then if i approved it being used as input.
safer than you'd think. also kind of annoying.

uneven pulsar
azure lynx
#

the 3 Pro is much better than the 2.5 Pro was

uneven pulsar
azure lynx
#

and only a 1x on the plan instead of 3x for Opus

#

i'm at 3/4 of my monthly usage.

uneven pulsar
azure lynx
#

Been busy. And had some thing get stuck for a while and kept saying "continue" without reading anything every time i looked back. that thinking probably took a lot.

#

eventually i just restarted in a new prompt and it did it in like a few minutes. it seems to work better if you make new conversations regularly.

#

ah.. only 1/3. was misreading it.

uneven pulsar
azure lynx
#

there is a sweet spot where it "understands" enough now to do the job but hasn't gotten distracted by all the other things it's done.

#

that's when it act's like a real person for a few interactions and you actually make progress.

uneven pulsar
#

Dam I don't have image sent perm

azure lynx
#

hmm. even if i ran out in co-pilot i've still got my daily uses with gemini-cli. just verified that seems to be working properly now with gemni 3.
one of my friends who still works at google swears by gemini-cli, but i think they get to use slightly better models than us mortals.

opaque sigil
#

the fact that gpt 5.3 codex costs as much as gemini pro is kinda of funny

fickle rain
#

OpenAI asking for too little or too much?

uneven pulsar
#

How is ure ai tranning was going?

#

They are offline :(

#

Ai tranning minimum ram 10 mb
But it smallest model

maiden geyser
#

insane wordington

true hemlock
#

the repair shop guy told me its a 3070

#

apparently it is a 3080Ti

b2be6b90-9deb-4cf1-879f-7a673d2c9675.jpg
uneven pulsar
azure lynx
#

I wonder if they drive JohnCar738 to JohnWork925

uneven pulsar
#

My mode tranning is getting cooked

#

num params: 6528
step 1 / 1000 | loss 4.7417
step 2 / 1000 | loss 4.6436
step 3 / 1000 | loss 4.5674
step 4 / 1000 | loss 4.3682
step 5 / 1000 | loss 4.2548
step 6 / 1000 | loss 4.6464
step 7 / 1000 | loss 4.3826
step 8 / 1000 | loss 3.9831
step 9 / 1000 | loss 4.1971
step 10 / 1000 | loss 4.2112
step 11 / 1000 | loss 4.1087
step 12 / 1000 | loss 3.8753
step 13 / 1000 | loss 3.5269
step 14 / 1000 | loss 4.2356
step 15 / 1000 | loss 4.7603
step 16 / 1000 | loss 2.8014
step 17 / 1000 | loss 3.5131
step 18 / 1000 | loss 3.0281
step 19 / 1000 | loss 3.5288
step 20 / 1000 | loss 2.6713
step 21 / 1000 | loss 3.6145
step 22 / 1000 | loss 6.5406

#

Well i put song lyrics 🤣

sage crag
uneven pulsar
#

And im waiting for output :)

#

And maybe take 12 mintus

azure lynx
#

i wish there was one standard URL and method for the "/ping" style health check endpoints across every service. ;/

#

i stupidly assumed they'd all be GET and now two of them need POST.

#

woo! now it can handle /health endpoints that require POST.

05FB2EA5-1C3F-4B6E-972D-F4BA7F1C0C60.png
#

and I get a green square!

#

LLMs are good at making repetitive edits to JSON files, like converting all the services from a "network" + [ paths to endpoints] -> [ "url" + "protocol" ] schema was trivial for it.

opaque wharf
#

Yeah, that's what they're good for when coding basically. Making boilerplate and throw away test/playground code

uneven pulsar
#

Wrong data :)
--- inference ---
sample 1: 11:04.390 --> 03
sample 2: 08:01.230 --> 00
sample 3: 04:05.090 --> 10
sample 4: 11:07.270 --> 05
sample 5: 04:26.980 --> 04
sample 6: 01:18.460 --> 13
sample 7: 03:07.210 --> 03
sample 8: 18:39.530 --> 13
sample 9: 07:24.580 --> 15
sample 10: 01:10.350 --> 03
sample 11: I ou tehyou tou
sample 12: 05:49.930 --> 10
sample 13: 01:24.460 --> 16
sample 14: I'l tou tonon d
sample 15: I tal the tou th
sample 16: I thish t thin s
sample 17: 02:18.980 --> 09
sample 18: It thin ane we s
sample 19: 01:08.290 --> 16
sample 20: 04:23.130 --> 11

azure lynx
#

it might need a little longer in the oven.

uneven pulsar
azure lynx
#

perhaps there might be some timestamps that need cleaning

uneven pulsar
#

I will be back

midnight sigil
uneven pulsar
#

Model was fine

midnight sigil
#

aight

uneven pulsar
#

step 1 / 1000 | loss 3.0075
step 2 / 1000 | loss 2.8725
step 3 / 1000 | loss 2.9672
step 4 / 1000 | loss 2.9229
step 5 / 1000 | loss 2.3033
step 6 / 1000 | loss 2.0312
step 7 / 1000 | loss 2.4992
step 8 / 1000 | loss 2.2207
step 9 / 1000 | loss 1.9362
step 10 / 1000 | loss 1.3447
step 11 / 1000 | loss 2.1302
step 12 / 1000 | loss 1.4773
step 13 / 1000 | loss 1.6167
step 14 / 1000 | loss 0.6918
step 15 / 1000 | loss 1.8763
step 16 / 1000 | loss 0.9096
step 17 / 1000 | loss 1.3725
step 18 / 1000 | loss 0.3497
step 19 / 1000 | loss 1.5835
step 20 / 1000 | loss 0.6389
step 21 / 1000 | loss 1.0829
step 22 / 1000 | loss 0.2662
step 23 / 1000 | loss 1.1805
step 24 / 1000 | loss 0.4923
step 25 / 1000 | loss 0.8260
step 26 / 1000 | loss 0.3201
step 27 / 1000 | loss 0.8148
step 28 / 1000 | loss 0.3806
step 29 / 1000 | loss 0.6392
step 30 / 1000 | loss 0.3219
step 31 / 1000 | loss 0.5905
step 32 / 1000 | loss 0.3412
step 33 / 1000 | loss 0.4923
step 34 / 1000 | loss 0.2496
step 35 / 1000 | loss 0.4713
step 36 / 1000 | loss 0.3295
step 37 / 1000 | loss 0.3778
step 38 / 1000 | loss 0.2513
step 39 / 1000 | loss 0.3933
step 40 / 1000 | loss 0.3016
step 41 / 1000 | loss 0.3188
step 42 / 1000 | loss 0.2783

midnight sigil
uneven pulsar
midnight sigil
uneven pulsar
#

step 438 / 1000 | loss 0.2231
step 439 / 1000 | loss 0.1640
step 440 / 1000 | loss 0.2324
step 441 / 1000 | loss 0.1499
step 442 / 1000 | loss 0.2230
step 443 / 1000 | loss 0.1639
step 444 / 1000 | loss 0.2324

#

And It's overfitting

midnight sigil
true hemlock
#

why did every 3080Ti i touched would always burn a single mosfet

#

this is the second time

opaque wharf
true hemlock
#

on the second 3080Ti i've touched

true hemlock
#

so uhhh

#

i got both good news and bad news

#

the bad news, already told

#

the good news though... i desoldered the inductor associated with that burnt mosfet and it seem like it got initialized

azure lynx
#

good news is you've got hot air rework tools and a replacement mosfet?

true hemlock
#

though my test bench is stuck on BOOT debug led

#

but at least its NOT "VGA"

#

so imma assume it actually gets initialized

uneven pulsar
azure lynx
#

interesting. seems like you can (according to some people on the internet) get by with dead VRM lines.

opaque wharf
true hemlock
opaque wharf
#

Also depends on how the power line is wired

true hemlock
#

removing one won't change much

azure lynx
#

will probably just stress a few others a little harder. make your super expensive GPU die a year or two earlier than otherwise

true hemlock
#

could always undervolt

#

also i got it for $14 so

#

oh

#

OH

#

WAIT

#

OH SHIT

#

a bit display just now

rigid snow
#

is this 100

#

it has to be no?

maiden geyser
#

are you 100vestigating rn

azure lynx
#

not a new account. months old. lots of details in it. probably not.

maiden geyser
azure lynx
#

do they have a history of buying other active accounts or do they just make a new one each time?

#

now i feel like it's working properly.

62D78C52-5EA7-49E4-AC52-12FB6891E846.png
umbral wigeon
#

Why C# eating my ram so hard

#

What is mircosoft visual studio doing

warped narwhal
opaque sigil
olive sable
#

vs2026 when ram exists NeuroNom

azure lynx
#

244 words in 6 seconds. except i didn't want it to wait until it had finished speaking so i need to redesign that part of the flow. it's streaming up until it hits a certain point. need to re-streamize it.
and if i'm getting that rate i can try bumping up the model size maybe.

#

(it's a problem because the TTS limits the input to 500 characters or it truncates it. ;/ and because 6 seconds is about 4 seconds too long. for a response.)

opaque wharf
olive sable
#

hdmi input software support is dogshit

opaque wharf
#

I wish I have opi5 to test it out too lol

#

I didn't even know they have HDMI in when I recommend it to you ngl

sage crag
fiery anchor
#

oh neat

opaque wharf
#

Man, the kernel maintainers sure is awesome

mighty thorn
rigid snow
#

rider has a ce now

uneven pulsar
#

Hello

#

How is u're life was going ?

#

BTW I JUST FINISHED MY LLM MODEL

#

This channel is dead
:(

olive sable
rough bloom
opaque sigil
opaque wharf
#

It's that isn't it?

olive sable
#

Engineering 100 ReallyInnocent

20260214_190531.jpg
fiery anchor
#

you'd be suprised how often you stee stuff like that in the wild.

olive sable
#

if it works it works

fiery anchor
#

ye

opaque wharf
sage crag
#

how are you

olive sable
tender river
opaque wharf
quick condor
sage crag
rigid snow
#

gah

opaque wharf
tender river
#

I'm getting to the point where I'm being confused by large numbers (almost running out of fingers and toes again), so the next kernel is going to be called 7.0

#

:mlntcandy:

rigid snow
#

lady awa of wrr is typing

sage crag
rigid snow
#

obsessed with numbers

#

send them into the arg thread

sage crag
sage crag
#

the minor version just refers to the merge window and the major version just refers to how many times the minor version got too big

rigid snow
sage crag
#

ye

#

why would you think i thought otherwise

rigid snow
#

because i thought you read it as “you’re just obsessed with numbers” which is dismissive of your point so you decided to defend it

sage crag
#

my literacy

#

insulted

#

make a rope out of grass and use it to drag a tonne for 10 laps

tender river
#

vedalPray please work this time

sage crag
#

no i dont wanna

#

work boring

tender river
quick condor
# sage crag

6.17 was terrible enough for me, had to revert to 6.11 due to instability

rigid snow
quick condor
#

To be fair though it was only partially the kernel, it was also the fact that I was doing P2P on 3090tis on an x299 chipset.
The way p2p is handled by x299 isn't great and it causes an obscure voltage sag on the cpu that I only found after fixing the parched driver and fixing the kernel version

real sierra
#

I always thought the Soviet water integrators were cooler tho

sage crag
#

shmo

real sierra
#

omg

#

koniiiiiiwrrrrrrrrbwaaaaa

olive sable
real sierra
sage crag
#

bweero

real sierra
#

shiro 📣 bweee bweeeeee bweeeeeeeeee

fiery anchor
sage crag
#

bweee

#

:shiro:

rigid snow
#

this channel is a psych ward and shiro is staff who’s also kinda chill

jagged turtle
sage crag
#

to prevent

#

mind reading

unkempt citrus
sage crag
opaque wharf
sage crag
#

i dont mind reading

unkempt citrus
#

I'd even go so far as to say I'm a big fan

olive sable
quasi sundial
#

kronii

olive sable
maiden geyser
umbral wigeon
#

Looks stable to me

warped narwhal
#

that pfp is incredible

rigid snow
#

it’s very 2014 if that makes sense

molten island
hasty plaza
#

hello, i got an idea to make a game but i know absolutely nothing of programming or anything like that, what program could i use to make a game with real time combat? 2d most likely

olive sable
#

unity or godot

#

there are also some more niche engines like rpgmaker for more specialized stuff

azure lynx
#

depends on how you want it to look. also "absolutely nothing" of programming is going to really make it hard.

olive sable
#

you're gonna have to program if you're making a game

hasty plaza
#

i know lol

#

rpgmaker can make real time combat?

olive sable
azure lynx
#

what do you mean by "real time combat"?
how do you want it to look and how do you want things to be controlled?
coz "real time combat" sounds very complicated and messy.

hasty plaza
#

not turn based +

fiery anchor
#

"real time combat" is a pretty broad term. from pong to E.V.E.

hasty plaza
#

i dont know how to say it

#

im ESL

azure lynx
#

what game is like it?

hasty plaza
#

recettear like something like that

rough bloom
fast pagoda
#

for the 2 people who will run pytorch via rocm on RDNA4 // specifically gfx1201 navi 48 (possibly, havent been able to test on my gfx1103yet) - dont set HSA_OVERRIDE_GFX_VERSION otherwise you'll be like me and spend a week with training runs doing Memory access fault by GPU node-* near step 0 over and over

rough bloom
hasty plaza
nocturne olive
fiery anchor
#

and coding isn't black magic. it is simply a different language to express yourself like english or math

nocturne olive